Abstract:Aiming at the problem of motor broken rotor bar fault diagnosis, a fault diagnosis model is designed. This paper transforms the rotor broken bar fault diagnosis problem into a classification problem, thus a fault diagnosis model based on multiple hyper-spheres support vector machine (MHSVM) is proposed. MHSVM is a multi-class classification model, which is constructed by using support vector data description (SVDD) and binary tree structure. To verify the effectiveness of the proposed algorithm, MHSVM is compared with support vector machine (SVM) and neural network algorithm (BP). The results show that the diagnosis rate of MHSVM is 94.92%, while the diagnosis rate of SVM model and BP model is 92.06% and 89.06% respectively. As can be seen, the diagnosis rate of the proposed model is the highest among the three models. The experimental results show that the effect of the fault diagnosis models based on SVMs is better than that of BP. Meanwhile, MHSVM has the best effect for broken rotor bar fault diagnosis, which proves the applicability and effectiveness of the MHSVM.

Abstract:In pulsed eddy current testing (PECT), the excitation parameters and sensor parameters have a great influence on the detection signal. Analyzing the influence law of different parameters on the detection signal can improve the performance of the detection system by optimizing the design. Based on the principle of pulsed eddy current testing, a test system for measuring the thickness of ferromagnetic specimen was established, and the influence of the number of turns and the inner diameter of the excitation coil on the test signal was analyzed. The test results show that increasing the number of turns and the inner diameter of the coil can improve the resolution of the detection signal. The increase of the inner diameter of the coil and the number of turns of the excitation coil can increase the order of magnitude of the minimum value of the detection signal. The detection thickness range of the sensor can be increased by increasing the number of coil turns and the inner diameter of coil.

Abstract:In order to solve the problem of environmental monitoring and fire extinguishing, this paper proposes a cooperative mode of the master control system and the control implementation system, which can collect, transmit and extinguish the data in the environment where the fire occurs. Using STC15F2K60S2 microprocessor and STM32F103C8T6 microprocessor as the main control system and control the implementation of the system control core, using Multi-sensor, amplifying circuit, wheel motor drive module, GSM and Bluetooth wireless communication module for data processing, temperature, smoke, flame and the human body detection, machine car perform obstacle avoidance logic of fire to track, and other functions. The experimental results show that the system can effectively monitor the environmental data and alarm when the threshold value is exceeded, and analyze and deal with the current environmental conditions. Realize the car search fire source and water function, to achieve the desired effect.

Abstract:In order to explore the influence of design parameters on the lightning low frequency magnetic field detection antenna, this paper designs a lightning low frequency magnetic field detection antenna. Firstly, based on the LCR analyzer, the influence of the antenna length, the number of the wire and the diameter of the magnetic rod on the resistance, capacitance and inductance of the antenna is analyzed. Secondly，a test platform was constructed using Helmholtz loop coils, and then the platform was used to perform sensitivity tests on the reception performance of ferrite antennas wound. And the effects of antenna length, the number of wires and the diameter of the magnetic rod on the receiving performance of the antenna were discussed. Finally, the research conclusions are as follows: Increasing antenna length can increase the antenna bandwidth, but it will reduce the frequency response value of the antenna. However, increasing the number of wrapped wires is the opposite. Increasing the magnetic rod diameter can increase the frequency response value and the bandwidth, but it will make the size of the antenna larger. So, the selection of the antenna length, the wires number and the magnetic rod diameter needs to be considered in a compromise. This paper analyzes the influence of various parameters on the lightning low-frequency antenna, which has a certain reference value for improving the lightning detection ability.

Abstract:The traditional method has the problem of large indication error in the range of 247mm-256mm from the measuring point. Therefore, a calibration method of measuring error of outer diameter micrometer based on Monte Carlo method is proposed. The UGS model of the outside micrometer is made and imported into the CAE software for processing, and the finite element analysis model of the outside micrometer is constructed. The vibration error model of measuring force change is constructed by Monte Carlo method, and the measurement error of external micrometer is analyzed by this model. Through the two gauge calibration compensation method, the measurement error of the external micrometer is compensated, so as to realize the error calibration. The experimental results show that the indication error of this method is smaller than other methods in the range of 247mm-256mm from the measuring point, which can effectively improve the error calibration effect.

Abstract:In order to improve the acceptance level of wind power and solve the problem of automatic generation control included in wind power, by combining with the characteristic of large-scale wind farms group access to grid section, the paper formulates wind power cluster section active power control strategy, and amends its ctive power control delay based on active response characteristic, gets the calculate method of section margin. Then, it builds fracture surface self-adaptation control frame that facing cluster wind based on Cyber-Physical Systems framework, and complish the optimizing of active power control strategy based on dynamic section margin calculation structure. Simulation results show that, the design method in this paper can reduce the wind turbine wear and operation cost, provide a guarantee for the wind farm output regulation rate, and reduce the output overregulation at low wind speed.In addition, the capacity of wind power can be fully utilized to reduce the line loss of cluster active power, and the economic benefit is obvious.

Abstract:In order to better evaluate the building quality and guarantee the building safety, this paper studies the detection of reinforcement in concrete to realize the accurate detection of the thickness of the protective layer and diameter of reinforcement without prior information. Ultra wideband stepped frequency ground penetrating radar and electromagnetic induction detection modules are developed. A relative permittivity estimation method is proposed, and the accurate thickness of protective layer can be obtained by combining radar data. The electromagnetic induction calibration experiments with different diameters and thickness of reinforcement were carried out, and the calibration database was obtained. The measured electromagnetic induction data and the thickness of protective layer are matched with the calibration database to obtain the exact diameter of reinforcement. The experiment on concrete test block shows that the steel bar can be detected without prior information, and the relative error of the thickness of the protective layer is less than 2%, and the diameter of the steel bar is detected without error. The system and method are proved to be effective and accurate, and can provide effective technical support for building quality evaluation.

Abstract:In order to eliminate UAV potential safety hazard in expressway patrol, and avoid the crash risk because of collision, the paper uses 3D dynamic collision prediction and optimal problem dynamic collision avoidance arithmetic to propose an UAV dynamic collision avoidance technology that is appropriate for three-dimensional space. And then, it uses 5G communication technology to construct communication network, when UAV meet random obstacle, it can make it maintain smallest safety clearance. Simulation results show that, paper’s technology is better than the technology based on optimal geometric in algorithm efficiency, route cost and route quality 3 aspects, the collision avoidance strategy is useful.

Abstract:To address problems of logistic robots, including the difficulty in positioning and posturing, poor interaction, monotony of tasks (carry and capture), etc., an intelligent logistic robot is designed. The robot is assigned with a task from the server via DT-06WIFI or through QR code scanning at Openmv and then informed of the position of the object by identifying bar code. The robot is arranged with a CN-TTS voice module and OLED display for reporting the robot's exceptional information and showing its normal information respectively. The robot enables capture in three dimensions based on parallel four-link mechanism and multi-link mechanism and adopts modified dual-loop (angle and speed) serial algorithm and chassis control strategy of kinematic resolution module. Through actual tests, the design scheme can realize a series of logistics handling tasks such as positioning, walking, recognition and grasp, etc. The design of Multi-sensor fusion make the robot applicable to a broader range of scenarios and can carry out tasks as required with much better human-machine interaction.

Abstract:Based on Michelson interference, we propose a micro-displacement information extraction method using imaging principle and image processing. The processing object of the method is equal inclination fringes collected by planar-array CCD camera. The image preprocessing for each frame includes denoising, top-hat transformation, binarization, morphology operations. Then radius of the bright interference fringe at the center of the image is obtained by feature recognition. By comparing the change trend and amplitude of the radius of frame sequence, the method realizes displacement information extraction. Experiments show that detection resolution of the method is λ/4 by making use of the information provided by every frame, which is superior to the maximum resolution of λ/2 of the detection by interference fringe counting .

Abstract:In view of the high equipment cost and long training time required by common deep learning methods for fault diagnosis, this paper proposes a bearing fault classification method based on the Inception-ResNet model. By using the Inception network's parallel structure, the network learns features of different scales, resizing structures are introduced to reduce degradation caused by network deepening, and three-dimensional convolution is added to allow information between different channels to blend. In order to verify the performance of this method, case Western Reserve University data set and IMS data set were used for verification, and compared with the traditional shallow learning method and deep learning method, experiments were conducted. The results show that, compared with other methods, the method presented in this paper not only has excellent diagnostic ability, but also is better in terms of resource utilization and training efficiency.

Abstract:Ultrasonic flowmeter has a good application prospect in the field of industrial production because of its good site applicability. In order to meet the requirements of the actual production for the accuracy of liquid flow measurement, the cross-correlation algorithm is used to detect the ultrasonic transit time (TOF), which reduces the calculation time error. Using FPGA internal resources to design and implement cross-correlation algorithm module, and adjust the logical structure to simplify the algorithm, reduce the resource duty cycle. The test results show that this method can effectively improve the measurement accuracy of transit time. In the range of 0-3000mm, the relative error of transit time is as low as 0.44%. The relative error of ultrasonic flowmeter is less than 3.68% in laminar flow region and less than 0.64% in turbulent flow region.

Abstract:Basic oxygen furnace (BOF) steelmaking plays an important role in the steelmaking process, so it is very important to control the converter process accurately. The control process of BOF steelmaking is based on the end point prediction of BOF. Therefore, in order to achieve accurate steelmaking, it is necessary to establish the end point prediction model of BOF steelmaking. According to the characteristics of BOF end-point prediction, an improved fuzzy support vector regression (IFSVT) algorithm is proposed in this paper. IFSVR is built by introducing the parameter based on fuzzy support vector regression (FSVR). In addition, in order to improve the optimization efficiency, particle swarm optimization (PSO) algorithm is used in the parameter optimization of IFSVR, so as to improve the modeling speed. The experimental results show that the proposed models are effective and feasible. Within different error bounds (0.005% for carbon content model and 10 C for temperature model), the hit rates of carbon content and temperature realize 91% and 94%, respectively. And a double hit rate of 90% is obtained, which can provide a significant reference for real BOF applications, and the proposed method is also appropriate for the prediction models of other metallurgical applications.

Abstract:A new low-complexity deep learning method is proposed to improve the complex neural network belief propagation algorithm, and reduce the complexity of the hardware implementation. By giving an alternative graph of the confidence propagation decoding algorithm and combining with the min-sum algorithm, the hyperbolic function operation is removed and the multiplication operation which consumes a lot of resources in engineering practice is converted into a simple addition operation. And combined with the recurrent neural network, different parameters of multiple layers are constrained into single layer parameters. Considering the attributes of different information, no additional parameters are attached to the log-likelihood ratio message from the channel, parameters are attached to the edge when the check node is updated. Combined with the distribution of edge parameters, it is found that part of weight values deviate greatly from 1, which is determined as the edge that needs to add weight. parameters of neural network are reduced by about 20%. performance of the proposed algorithm is improved by 1dB compared with the belief propagation algorithm in the high SNR regime, which is convenient for hardware implementation and has strong practicability.

Abstract:This paper studies the influence of the ratio of main current & bucking current and the rest potential of the hard focus lateral tool on the logging results, the circuit simplified measurement method of the the ratio of main current & bucking current and the rest potential, simplifies the circuit model of the main current & bucking current measurement of the tool with the help of the forward method, builds the main current test circuit and the bucking current measurement circuit, and gives the theoretical calculation method and the result analysis. The results show that the method can be based on the electricity simplified model, the ratio of main current & bucking current and rest potential test results are given. The test results can correct the errors of the forward model of the tool to a certain extent. The application advantages of this method in the main focus circuit parameter test are analyzed. It can quantize and compare the parameters of the main current measurement in different directions of the array azimuthal laterolog, ensure the consistency of the tool orientation, and provide the data support for the continuous optimization of the main focus circuit.

Abstract:To solve the problems of storing hardware and volume of database limitation in the archiving and storing scenario of massive well logging data, a lossless data compression method based on deep learning is proposed in this paper. Data stream is compressed by adaptive arithmetic encoder combining current byte with outputted conditional probability distribution of data stream by recurrent neural network RNN as probability predictor. Data stream is decompressed by saved weights of RNN and arithmetic decoder. Compared with traditional lossless compression methods, compression ratio of one dimensional log data is improved by about 23% averagely and that of two dimensional array log data is improved by about 21% averagely in actual compression test. At the same time, A large scale well logging data storing database constructing method based on multi dimensional featuring indexes querying tree structure is proposed combined with lossless compression method in this paper whose querying efficiency is improved by about 45% compared with traditional database querying method under the multi conditions combining query. Results show that storing space of logging data is decreased effectively with lower data querying time by the method in this paper which provides technical base for storing and utilization of large volume logging data and saves hardware cost and labor cost of data archiving.

Abstract:In this paper, an active color control method is proposed to detect the scanned image of FPD system. The brightness of the light source is adjusted by using the high-power light-emitting diode (LED) and communication port of the mixed color source. The mixed light irradiates the active matrix organic light-emitting diode (AMOLED) panel after passing through the beam expander and forming a harness. After intensity correction, the image quality is evaluated by Tenenbaum gradient, and the dimming level is determined by simplex search method (SSM). Each time the AMOLED panel is scanned, the color of the light will change and will be repeated until the image quality is close to the local maximum value. The number of scans performed is less than 225, and the number of dimming level combinations is 20484. The proposed method reduces the manual setting of detection equipment and realizes automatic color lighting for obtaining high-quality images. It has certain practical value for FPD detection equipment.

Abstract:To solve the problem of low classification accuracy and high computational cost in the qualitative data environment, a classification variable identification method was proposed to improve the classification separability by using traditional classifiers and different mapping techniques. By mapping the initial feature (classification attribute) to the real domain space and using the chi-square (C-S) as the measure of difference, the dimension of the feature space is increased to improve the class separability. The t-distributed domain embedding algorithm (tSNE) is used to reduce the dimension of the data to two or three features, thus reducing the calculation time of the learning method. It is proved by experiments on the common classification data set that C-S mapping and t-SNE not only guarantee the recognition accuracy, but also greatly reduce the computation of recognition task. At the same time, when only C-S mapping is applied to the data set, the separability of categories is enhanced, thus significantly improving the performance of the learning algorithm.

Abstract:Because the infrared imaging system will be disturbed by various kinds of external environment in the signal transmission and signal conversion, the infrared imaging system will produce many kinds of noise in the infrared image generated, which will lead to the decrease of the signal-to-noise ratio of the infrared image.In view of the problem that infrared images are affected by multiple types of noise,to make a new weighted coefficient based on the iteratively reweighted least squares algorithm of compressed sensing theory and combined the principle of median filtering to construction a new reconstruction algorithm.Firstly, the median filter is used to do rough denoising of infrared image.Then fine denoising by compressing the sensing sparse transformation and observation matrix to keeping the observations important information about the original signal.Finally, the denoised by reconstruction algorithm.The algorithm is simulated in MATLAB to verify the effectiveness of the algorithm.Experiments show that the visual effect of the image obtained by this algorithm is close to the original image, and it has better denoising performance in the actual scene. The peak signal-to-noise ratio is 3dB～8dB higher than the original SP algorithm and 5dB higher than the iterative weighted least square algorithm.5dB ~13dB.

Abstract:Aiming at the limitation of a single sensor in intelligent vehicle target detection, a target detection algorithm based on four-layer lidar and camera fusion was proposed. The position and number information of the target is obtained by lidar, and the result obtained after point cloud clustering is projected onto the image through the parameter matrix of joint calibrated by lidar and camera to obtain the bounding box of the target. The boundary box, category and confidence of the target are obtained by the collected images through the yolov3 network. Then, the decision level fusion method is used to fuse the detection results of lidar and camera, and the final detection result is obtained. The experimental results show that the detection rate of vehicle and pedestrian is 94.8% and 96.4% respectively. Compared with other methods, this algorithm can improve the detection rate and robustness.

Abstract:This paper studies a fuzzy neural network algorithm which can effectively promote the efficiency of urban rail transit operation organization and management. The overrun learning machine module is used to convolute the passenger flow of each station, the convolution neural network is used to summarize the passenger flow data of each station in the inspection line, and the reference data of other lines is constructed, and the binary module is used to form the release signal light recommendation data. After the application of the system, the peak vehicle load rate of passenger flow is significantly reduced, the estimated vehicle load rate of passenger flow is significantly increased, and the departure gap of passenger flow is significantly increased, but it does not affect the residence time of passengers in the station. It is considered that the algorithm can effectively improve the operation efficiency and economic benefits of urban rail transit.

Abstract:Aiming at the problem that most of the existing smoke detection methods are only suitable for the environment with sufficient light and the detection effect is poor in the low light environment, this paper proposes a low light smoke detection method based on FSSD. Firstly, the video frames containing moving objects are extracted based on single gaussian modeling method; secondly, the low contrast and low signal-to-noise ratio characteristics of low light level smoke image will cause difficulties in target detection, so the image preprocessing method combining the contrast limited adaptive histgram equalization and median filter is designed. Finally, in order to strengthen the early warning ability, the FSSD network that is conducive to detecting small targets is adopted, convolutional block attention module is embedded in the output of the network, which can strengthen the important feature information and improves the accuracy of target detection. On the low light smoke dataset, the Recall, Precision, and F1 of the proposed method reached 97.5%, 93.3% and 95.4%, indicating that the method is effective and can be applied to smoke detection in low light environments.

Abstract:Osteoarthritis is the most common joint disease in middle-aged and elderly people. The disease and its complications account for 10 per cent of global medical problems. Among them, Osteoarthritis of the knee is the most serious and the risk of disability is very high. Early detection and interventional treatment is of great significance to relieve the symptoms and reduce the harm. In this paper, a large number of knee joint DR image data were collected. Various texture features and fusion features were extracted from the obtained data. Then, various combinations of extracted feature vectors were used as input to the training support vector machine model. We use the grid search method to optimize the parameters. The highest accuracy of the trained model on the test set can reach 84.29%, which has good intelligent classification and diagnosis performance. Using the trained SVM model can effectively grade knee osteoarthritis and assist doctors in diagnosis, which is of great significance for early diagnosis and early intervention treatment of knee osteoarthritis.

Abstract:In order to solve the problem that the link prediction method based on single relational path cannot mine the influence of different paths in the knowledge map, a link prediction method based on multi relational path is proposed. Firstly, the similarity index based on path information is used to calculate the similarity between all relational paths. Then, the relationship projection between different paths is extended to the new path projection and path constraints, and the training process is performed by using random gradient descent, so that the explicit features between different paths can be screened out through low dimensional representation learning in implicit space. The validation analysis is carried out on Enron email data set and National Natural Science Foundation data set. Experimental results show that, compared with other path link prediction algorithms, the maximum improvement of map and AUC is about 20%, showing higher prediction accuracy.

Abstract:In the wireless ultraviolet communication system, the use of appropriate modulation and demodulation technology can effectively improve the system communication quality. In this paper, a fixed-length dual-pulse symmetrical pulse interval modulation (FDPSPIM) method is designed to realize ultraviolet communication, and its symbol structure is analyzed in detail. The modelsimSE 10.1c is used for timing simulation to verify the feasibility of the modulation scheme. At the same time, the hardware description language is used The FDPSPIM modulator and demodulator design based on Field Programmable Gate Array (FPGA) realizes the synchronization processing of sending information through a custom frame structure. The test results show that: based on the built-up ultraviolet light communication experimental platform, the FDPSPIM modulator and demodulator designed by Verilog HDL can correctly identify the original information.

Abstract:In order to overcome the large size, high power consumption and low radiation efficiency of the transmitting antenna in the existing low-frequency communication system, a miniaturized and low-power low-frequency transmitting antenna is proposed. The low frequency transmitting antenna uses servo AC motor to drive the permanent magnet to rotate to generate ultra-low frequency electromagnetic wave. The upper computer sets the motion parameters of the servo motor through the programming software of motion controller, and sends them to the controller to control the motor motion in real time. The spatial modulation technology of FSK is applied to the mechanical antenna for information transmission. The experimental results show that the mechanical antenna can achieve 1bps ultra-low frequency communication with a 27.7 cm3 radiation source of permanent magnet and a low power consumption of 44W. Compared with traditional low-frequency transmitting antennas, the mechanical antenna is small in size, low in power consumption, and radiation efficiency is not limited by the Chu-Harrington limit, providing a new technical proposal for transmitting antenna of ultra-low frequency communication systems.

Abstract:The time spectrum is easily affected by the stratigraphic environment during the logging process, so it is necessary to filter out the noise in the time spectrum based on the energy spectrum information. For these requirements, a multi-channel energy spectrum acquisition and processing circuit was designed. This circuit used two STM32 as the main controller. Based on CFD (Constant Fraction Discriminator), the circuit adopted the trailing edge of the original narrow pulse as the peak detection moment, and used external trigger ADC to collect the peak value of the pulse signal. It obtained the time spectrum and energy spectrum information of the pulse signal generated by the three probes at the same time, and communicated with the ground system through the CAN bus. The circuit has the characteristics of small size, low power consumption, and high real-time performance. Experimental results show that the circuit can accurately collect energy spectrum information and meet the needs of O/A tool.

Abstract:In order to further analyze and optimize the performance of capacitive ultrasonic transducer (CMUT), the equivalent circuit models of CMUT cell and CMUT array were established, and the equivalent circuit simulation was carried out. First, CMUT small signal equivalent circuit model was deduced, and then the circuit simulation software Advanced Design System (ADS) was used to establish CMUT cell equivalent circuit for the calculation. The influence of different amplitudes and periods of AC voltage on the emission characteristics of CMUT cell was analyzed. The results show that the larger the amplitude and duration of AC voltage, the larger the vibration displacement and output sound pressure of the CMUT cell, and the better the emission performance. In addition, the equivalent circuit model of CMUT array was derived based on the mutual radiation impedance, and the frequency characteristics of CMUT array with different cell sizes, spacing and numbers were analyzed by ADS simulation. The results show that the size, spacing and number of cells all affect the frequency of CMUT array. The above analysis results provide theoretical reference for the design and testing of CMUT.

Abstract:In order to solve the problem of artificial characteristics of bearing fault diagnosis, this paper puts forward the fault diagnosis method of the time frequency analysis and VGG19 network migration learning. First, the normal state, the internal ring fault, the outer ring fault and the sliding fault of the rolling body are converted to the frequency sample diagram, and then the spectrum kurtosis graph is generated from the above data. Secondly, the full connection layer in the VGG19 network model is replaced and fine-tuning. Finally, the convolutional neural transfer learning network is used to recognize and classify bearing faults through network parameter tuning.The results show that the classification accuracy of time-frequency sample graph for rolling bearing fault diagnosis in the experiment is 5.42% higher than that of spectral kurtosis graph, which verifies the validity of the application of time-frequency analysis and VGG19 transfer learning in signal processing.In addition,Transfer learning can solve the problem of fault diagnosis of small samples.

Abstract:Low concentration gas detection is a hot issue in gas detection. Hydrogen, as a combustible gas and diatomic molecule, has always been the focus of gas detection research. However, in the current detection methods, gas chromatography technology can be used for quantitative analysis or even trace detection, but the test system is complex and expensive; hydrogen electrode detection technology can be used for continuous in-situ detection, but the test range is large and the accuracy is not high; oxidation titration technology is easy to be detected The effect of reducing substances, such as oxygen, has a large error. In order to solve these problems, this paper uses NDIR technology to analyze the key factors that affect the precision measurement of hydrogen concentration detection, and optimizes the design. A high-precision hydrogen concentration measurement and control system is completed. Through experiments, the standard concentration of 2.54%, 6.35%, 8.00%, 11.25%, 20.00% sample gas is tested, and the absolute error of detection is controlled at 0.1 %In the following, the high-precision detection of low concentration hydrogen is realized. After long-term operation monitoring, the experimental equipment has good stability, which verifies the correctness and effectiveness of the improved design.

Abstract:Based on the research of excitation sensor based on transient electromagnetic detection technology, a kind of sensor with focusing magnetic field function is designed, which is mainly composed of double excitation coil, detection coil and magnetic core. The results show that the sensor mainly concentrates the magnetic field in the region of R = 250mm, and the magnetic induction intensity B = 1.745 × 10-5, which is 50% smaller than the single coil magnetic field concentration range, and the magnetic induction intensity is increased by 31%, which meets the design requirements, and the magnetic field distribution gradient is large, which is helpful for the later data processing Reason. The sensor can effectively detect the local corrosion of metal pipeline, provide key components for transient electromagnetic detection technology, and provide the basis for the structure optimization and field application of the sensor.

Abstract:Electromagnetic ultrasonic transducer can be used for non-contact detection, which is convenient to use, but its disadvantage is low efficiency of energy exchange.To increase the energy conversion efficiency of electromagnetic ultrasonic transducer, The paper analyzed the working principle of electromagnetic ultrasonic transducer.it also modeled and simulated the electromagnetic ultrasonic transducer by using COMSOL Multiphysics. And the influence of different placement modes and geometric size of the permanent magnet on the excitation Lorentz force of the transducer was analyzed. By orthogonal experiment , the geometric parameters of the butterfly coil are optimized to obtain the best efficiency of shear wave energy transfer, Finally, The thickness test of aluminum plate was carried out by using ultrasonic detection system. The results show that the shear wave excitation efficiency is higher when the permanent magnet is placed vertically, and the amplitude of the ultrasonic shear wave excited by the butterfly coil is 50% higher than that by the spiral coil.T he butterfly coil has better shear wave energy transfer efficiency than the spiral coil.